| name | security-backend |
| description | Audit the .NET backend (apps/api) for security issues against OWASP Top 10 mapped to .NET/Minimal API/EF Core (Npgsql), plus tenant isolation, secret handling, and EU data residency. Use when reviewing backend changes for security, or running a backend security audit. Preloaded by the security-auditor-backend agent. |
| allowed-tools | Read, Glob, Grep, Bash, Skill, mcp__serena__initial_instructions, mcp__serena__get_symbols_overview, mcp__serena__find_symbol, mcp__serena__find_referencing_symbols, mcp__serena__find_declaration, mcp__serena__find_implementations, mcp__serena__search_for_pattern, mcp__serena__get_diagnostics_for_file, mcp__serena__list_dir, mcp__serena__find_file |
Serena MCP is mandatory for code (read-only here). First call mcp__serena__initial_instructions to load the Serena tool manual, then use Serena for ALL code reading / searching / navigation — prefer symbol navigation (get_symbols_overview / find_symbol / find_referencing_symbols) over whole-file reads.
Backend security audit ({{ProductName}} .NET API)
The procedure for a backend security review. The full, cited check catalog is in
reference/owasp-dotnet-checks.md (OWASP A01–A10 mapped to our stack +
.NET caveats + a top-15 shortlist). This is backend-specific — the frontend has its own auditor.
Procedure
- Load the catalog — read
reference/owasp-dotnet-checks.md. It's the authority; don't re-derive checks.
- Scope — identify what changed / what to audit (a diff, a feature, or the whole
apps/api). Ground in the
actual code via Glob/Serena; the tenancy rule (.claude/rules/backend/tenancy.md) auto-loads here.
- Scan by category — walk A01→A10, flagging concrete violations with
grep/Serena evidence. Prioritize the
top-15 shortlist (blast-radius × likelihood) first.
- Complement, don't duplicate — items tagged
[OVERLAP] are already covered by our rules (tenancy /
persistence / result-and-errors / logging). For those, verify the rule actually holds at the security
boundary (e.g. a cross-tenant read, an overposting privilege escalation, an error-detail disclosure, a
secret/PII egress) rather than re-flagging the style concern.
- Classify — Critical (auth bypass / cross-tenant read / SQLi / RCE / secret exposure) · High · Medium.
Tenant-isolation and EU-residency failures are always Critical — they're the product's reason to exist.
- Report — per finding: file + symbol (+ line as a lead), severity, the OWASP category, the concrete
violation, the fix, and a reference URL from the catalog. Be specific and verifiable (findings get
adversarially checked).
Stack-specific hot spots (where to look first)
- Tenancy: every EF query /
FromSql / raw SQL carries a tenant_id predicate; global filter present;
IgnoreQueryFilters() audited; RLS as backstop. Writes set tenant_id in the domain factory.
- Authz: every Minimal API route
RequireAuthorization() / deny-by-default fallback; IDOR on id lookups.
- Injection:
FromSqlInterpolated not FromSqlRaw+string; EF 10 raw-SQL analyzer on.
- Overposting: bind to DTOs, never entities.
- Errors: ProblemDetails only; no stack traces;
EnableSensitiveDataLogging/EnableDetailedErrors Dev-only.
- Egress (SSRF + residency):
IHttpClientFactory, allowlisted EU-region outbound to model providers/pipeline.
- Secrets/PII: none in
appsettings/source/logs/OTel; JWT fully validated (CA5404); cookie flags set.
- Deps: NuGetAudit incl. transitive; any new package is a Dan-approval decision (no-library rule).
